AUTHOR=Tang Nan , Liu Shuang , Li Kangming , Zhou Qiang , Dai Yanan , Sun Huamei , Zhang Qingdui , Hao Ji , Qi Chunmei 

TITLE=Prediction of in-hospital mortality risk for patients with acute ST-elevation myocardial infarction after primary PCI based on predictors selected by GRACE score and two feature selection methods

JOURNAL=Frontiers in Cardiovascular Medicine

VOLUME=Volume 11 - 2024

YEAR=2024

URL=https://www.frontiersin.org/journals/cardiovascular-medicine/articles/10.3389/fcvm.2024.1419551

DOI=10.3389/fcvm.2024.1419551

ISSN=2297-055X

ABSTRACT=Precise prediction of in-hospital mortality subsequent to percutaneous coronary intervention is essential in informing and guiding clinical decision-making. Data Mining and Machine Learning methods have emerged as optimal estimators in medicine over the past decades. In this study, a dataset of 4677 patients related to the Regional Vascular Center of Primorsky Regional Clinical Hospital No. 1 in Vladivostok from 2015 to 2021 was extracted for in-hospital mortality prediction for patients with acute ST-elevation myocardial infarction after primary percutaneous coronary interventions. Extreme Gradient Boosting, Histogram Gradient Boosting, Light Gradient Boosting, and stochastic gradient boosting methods were selected for prediction tasks utilizing all datasets, and Monte Carlo Cross-validation was utilized for model selection. For the most relevant feature selection, the Recursive Feature Elimination method removed less important features, retraining the selected boosting models at each step until a complete ranking is obtained. Also, Shapley Additive Explanations ranked factors based on in-hospital mortality prediction based on selected models. Then, hybrid forms of models were developed utilizing four metaheuristic algorithms: Augmented Grey Wolf Optimizer (AGWO), Bald eagle search optimization (BES), Golden jackal optimizer (GJO), and Puma optimizer (PO) based on selected features by various methods and traditionally used factors (GRACE score). Prediction accuracy results introduced the best estimators to be used in hospitals instead of GRACE score calculated through online calculators (on medical websites) or some electronic health recorders. The findings revealed that in scenario (1), using GRACE scale features, the LGBM and XGBC models optimized using the BES optimizer achieve Recall values of 0.944 and 0.954, respectively, for accurate predictions, furthermore, in scenarios (2) and (3), where SHAP and RFE-selected criteria employed, the LGBE model attained Recall values of 0.963 and 0.977. In contrast, the XGBE model achieved values of 0.978 and 0.99, respectively.